1) I am not sure about EViews. To my understanding, Block exogeneity is related to colinearity and used for excluding the colinear variables. If you concern about non-stationality, 1st difference is the standard choice (note that it does work for non-stationality of the mean (average value) but not necessarily for non-stationality of the variance)

2) I can not catch the meaning, sorry

3) lag length can be at least AIC-type model selection. but as you may know, the sampling rate of the data is the same or much important than lag length. we should carefully resample the data (often we should not use the original sampling rate because of too much sampling for AR or VAR). lag length and sampling rate are interrelated and we should decide as the spectrum is represented properly (not too rough, not too detailed) when you calculated spectrum via AR model.

Dear Shuntaro Okazaki

My question is that, if there is non-stationarity is time series data, then

1)can we use Granger-causality at levels?

2) or should we use Granger Causality test at first differences?

as in many research papers, i found that researchers are using Granger causality at levels, which is a big violation of basic assumption of Granger causality test that series should be stationary.

thank you.

Dear Devesh.

I feel you should use the two and clarify the results. Then report the best out of the two and support it with literatures.

Regards

Regards